Create readme.md

Added the readme File

Author: KaiAllAlone

models/ML_Models/ResNET+CNN (Tranfer Learning)/readme.md

@@ -0,0 +1,169 @@
+# Eye Drowsiness Detection using ResNet50
+
+A deep learning project for detecting driver drowsiness by classifying eye states as "awake" or "sleepy" using transfer learning with ResNet50.
+
+## 🎯 Project Overview
+
+This project implements a binary classification model to detect drowsiness in drivers by analyzing eye images. The model achieves **98.5% validation accuracy** using a pre-trained ResNet50 backbone with custom classification layers.
+
+### Key Features
+- **Transfer Learning**: Utilizes pre-trained ResNet50 from ImageNet
+- **Binary Classification**: Classifies eyes as "awake" or "sleepy"
+- **High Accuracy**: Achieves 98.5% validation accuracy
+- **GPU Optimized**: Configured for Google Colab with GPU acceleration
+- **Production Ready**: Saved model ready for deployment
+
+## 📊 Model Performance
+
+| Metric | Training | Validation |
+|--------|----------|------------|
+| **Accuracy** | 97.76% | 98.50% |
+| **Loss** | 0.0600 | 0.0415 |
+
+## 🗂️ Dataset
+
+The project uses the **MRL Eye Dataset** from Kaggle containing:
+- **Training Set**: 50,937 images (80/20 split for train/validation)
+- **Validation Set**: 16,980 images  
+- **Classes**: 2 (awake, sleepy)
+- **Image Size**: 224×224 pixels
+
+**Dataset Source**: [MRL Eye Dataset on Kaggle](https://www.kaggle.com/datasets/akashshingha850/mrl-eye-dataset)
+
+## 🏗️ Model Architecture
+
+```
+ResNet50 (Pre-trained, Frozen)
+    ↓
+GlobalAveragePooling2D
+    ↓
+Dense(128, activation='relu')
+    ↓
+Dropout(0.3)
+    ↓
+Dense(2, activation='softmax')
+```
+
+### Model Specifications
+- **Base Model**: ResNet50 (ImageNet pre-trained, frozen)
+- **Input Shape**: (224, 224, 3)
+- **Output**: 2 classes (awake/sleepy)
+- **Total Parameters**: ~25M (only ~260K trainable)
+- **Optimizer**: Adam
+- **Loss Function**: Binary Cross-Entropy
+
+## 🚀 Getting Started
+
+### Prerequisites
+- Python 3.7+
+- TensorFlow 2.x
+- Google Colab (recommended) or local GPU environment
+- Kaggle API credentials
+
+### Installation & Setup
+
+1. **Clone the repository**:
+```bash
+git clone https://github.com/yourusername/eye-drowsiness-detection.git
+cd eye-drowsiness-detection
+```
+
+2. **Install dependencies**:
+```bash
+pip install tensorflow matplotlib kagglehub
+```
+
+3. **Run in Google Colab**:
+   - Upload `Drowsiness.ipynb` to Google Colab
+   - Enable GPU: Runtime → Change runtime type → Hardware accelerator → GPU
+   - Run all cells
+
+### Usage
+
+1. **Download Dataset**:
+```python
+import kagglehub
+path = kagglehub.dataset_download("akashshingha850/mrl-eye-dataset")
+```
+
+2. **Train the Model**:
+```python
+# Configure paths
+train_folder_path = '/path/to/train'
+val_folder_path = '/path/to/val'
+
+# Train model (5 epochs)
+history = model.fit(train_ds, validation_data=val_ds, epochs=5)
+```
+
+3. **Save the Model**:
+```python
+model.save('/content/saved_model/resnet.keras')
+```
+
+4. **Load and Use for Prediction**:
+```python
+import tensorflow as tf
+loaded_model = tf.keras.models.load_model('/path/to/resnet.keras')
+prediction = loaded_model.predict(new_image)
+```
+## 🔬 Training Details
+
+### Hyperparameters
+- **Image Size**: 224×224 pixels
+- **Batch Size**: 16
+- **Epochs**: 5
+- **Learning Rate**: Adam default (0.001)
+- **Validation Split**: 20%
+- **Data Augmentation**: None (can be added for improvement)
+
+### Training Results
+```
+Epoch 1/5: val_accuracy: 0.9753
+Epoch 2/5: val_accuracy: 0.9756  
+Epoch 3/5: val_accuracy: 0.9761
+Epoch 4/5: val_accuracy: 0.9806
+Epoch 5/5: val_accuracy: 0.9850
+```
+## 🔧 Customization & Improvements
+
+### Potential Enhancements
+- **Data Augmentation**: Add rotation, brightness, contrast adjustments
+- **Fine-tuning**: Unfreeze top layers of ResNet50 for better accuracy
+- **Real-time Processing**: Optimize for webcam/camera input
+- **Multi-class**: Extend to detect different levels of drowsiness
+- **Ensemble Methods**: Combine multiple models for better performance
+
+### Fine-tuning Example
+```python
+# Unfreeze top layers for fine-tuning
+base_model.trainable = True
+for layer in base_model.layers[:-10]:
+    layer.trainable = False
+
+# Use lower learning rate
+model.compile(optimizer=tf.keras.optimizers.Adam(1e-5),
+              loss='categorical_crossentropy',
+              metrics=['accuracy'])
+```
+
+## 📈 Performance Analysis
+
+The model shows excellent performance with:
+- **No Overfitting**: Validation accuracy consistently higher than training
+- **Quick Convergence**: Reaches high accuracy within 5 epochs
+- **Stable Training**: Consistent improvement across epochs
+
+## 🤝 Contributing
+
+Contributions are welcome! Please feel free to:
+- Report bugs or issues
+- Suggest new features or improvements
+- Submit pull requests
+- Share your results and modifications
+
+## 🙏 Acknowledgments
+
+- **Dataset**: MRL Eye Dataset by akashshingha850 on Kaggle
+- **Base Model**: ResNet50 from TensorFlow/Keras
+- **Platform**: Google Colab for providing free GPU resources